Introduction:

Multi-hearth furnaces are a key stage in the nickel reduction process. The implementation of the automatic control system in the furnace post-combustion has been affected by the lack of mathematical models of this process.

Objectives:

To obtain linear models for different operating points and non-linear models based on artificial neural networks that reflect the dynamic characteristics of the process.

Methods:

Carrying out active experiments, with pseudo-random binary sequences modulated in amplitude and frequency and inserted into the automaton that activates the air flow regulating valves and variations in the flow of mineral fed, to obtain linear models. In addition, they were recorded passive experiments of 10 months of operations to obtain neuronal models. It was evaluated the multiple-input-multiple-output neuronal model with different numbers of neurons in the hidden layer and with the use of the random cross-validation method, choosing the best model based on the Akaike and Bayesian information criteria.

Results:

The neural model predicts the temperatures of furnace hearths four and six with an error of less than 5°C, and a prediction horizon of one step ahead (120s).

Conclusions:

The model contributes to predicting the thermal profile in the heating zone of the furnace, as a basis for the design of control strategies that guarantee better use of energy and the reducing additive fuel, to reduce process losses and pollution environmental.

        · resumen en Español     · texto en Español     · Español ( pdf )